AI Reading for Wednesday January 29
Oh no OpenAI is big mad someone is taking advantage of OpenAI's IP to train AI models … pot kettle black much? - The Guardian
1,156 Questions Censored by DeepSeek - promptfoo
US firms are chomping at the bit to use DeepSeek - WSJ
New study suggest frequent use of AI makes us stupid. - UNILAD Tech
Using deep sea robots and crowd-sourced identifications to find new species and train AI - Bloomberg
Train in a perfect environment or a noisy one? - MIT News | Massachusetts Institute of Technology
AI model simulates evolution to generate a new glowing molecule - livescience.com
Alibaba releases new Qwen model. ‘Surpasses DeepSeek’ is the new 'Surpasses OpenAI'? - Reuters
Qwen is highly ranked against automated benchmarks (possibly overfitted), a little better than Llama on human evals. Will be interesting to see if it improves significantly and challenges the DeepSeek models which beat Llama by a fair amount. As a reminder, DeepSeek R1 reasoning model matches OpenAI o1 but is open source and costs a lot less, DeepSeek V3 base model beats Llama and other open-source models but doesn’t challenge GPT-4o and Gemini 2.0.
OpenAI safety researcher who left in November says he's terrified - X (formerly Twitter)
LinkedIn Removes Accounts of AI 'Co-Workers' Looking for Jobs - 404 Media
Vatican Warns About the Risks of A.I. - NY Times
Beware of forbidden knowledge, false AI-dols, byte of the serpent
'Companion' thriller showcases possible downsides of sexy robots - AP News
Should have paid attention to Pope Francis. Every day we stray further from God’s light.
Follow the latest AI headlines via SkynetAndChill.com on Bluesky